Sunday Rundown #89: New LLMs & Mortal Veggies
Sunday Bonus #49: Custom GPT for finding relevant Why Try AI entries.
Happy Sunday, friends!
Welcome back to the weekly look at generative AI that covers the following:
Sunday Rundown (free): this week’s AI news + a fun AI fail.
Sunday Bonus (paid): an exclusive segment for my paid subscribers.
Let’s get to it.
🗞️ AI news
This was one hell of a busy week!
👩💻 AI releases
New stuff you can try right now:
Alibaba Group open-sourced Wan2.1, a state-of-the-art AI video model that can render realistic motion and generate sound effects. (Try it on Hugging Face.)
Anthropic launched Claude 3.7 Sonnet, a top-tier hybrid model that combines traditional LLM and reasoning in one. (But it still lacks web access.)
ElevenLabs released Scribe, a speech-to-text model that understands 99 languages and boasts the industry’s lowest word error rate. (Try it for free.)
Google news:
Conversation branching in AI Studio now lets you explore different chat pathways while keeping the “parent” chat intact.
Gemini came to Google Sheets and can extract insights from data and generate visualizations like heatmaps.
Video transcripts in Drive help users easily search and find key moments in their videos via auto-generated transcripts for their videos.
Hume AI released an intelligent text-to-speech model called Octave that actually understands what it’s saying and accurately expresses emotions.
IBM open-sourced an experimental model family called Granite 3.2 that uses chain-of-thought reasoning to match the performance of much larger models.
Ideogram released Ideogram 2a which creates images faster and cheaper without sacrificing quality, optimized for graphic design and photography.
Inception Labs launched a diffusion large language model, Mercury, that matches the performance of other speed-focused models while being 10x faster.
Luma AI incorporated audio generation into its Dream Machine, letting you add AI-generated sound to your video creations.
Microsoft news:
You can now use Voice (voice chat) and Think Deeper (reasoning) features in Copilot for free!
Phi-4-multimodal and Phi-4-mini are the latest small-but-capable models for multimodal use cases and text-based tasks, respectively.
Midjourney lets you better organize your work with new folder tools and other quality-of-life improvements.
OpenAI news:
The Advanced Voice mode, powered by GPT-4o mini, is now available to ChatGPT free users as a preview.
Deep Research rolled out to all paid ChatGPT account types. Non-Pro accounts get 10 research queries per month.
GPT-4.5 is the company’s largest “traditional” LLM. It has better vibes (a la Claude 3.5 Sonnet) and is better at writing and conversation tasks. Out as a research preview for Pro users and developers, releasing to others soon.
Sora is finally available in Europe: EU, UK, Switzerland, Norway, Liechtenstein, and Iceland. (I’ve been using it for a while via VPN though.)
Perplexity brought a new voice mode to its iOS app, letting users ask questions and get spoken, real-time answers. (Coming to Android and Mac later.)
Pika’s new Pika 2.2 model can generate 10-second clips at 1080p and lets you use “Pikaframes” for seamless keyframe transitions.
Poe launched Poe Apps that let people use AI to code and deploy apps built on existing Poe models.
🔬 AI research
Cool stuff you might get to try one day:
Alibaba launched a preview version of its upcoming QwQ-Max reasoning model that should rival existing offerings by other AI labs.
Amazon announced a new, smart, AI-powered Alexa+. It will be free for Amazon Prime members and will start rolling out in the US soon.
Meta reportedly plans to launch a standalone Meta AI app for its open-source models to compete with offerings like ChatGPT.
Perplexity is working on a browser for agentic search called Comet. (Sign up for the waitlist.)
📖 AI resources
Helpful AI tools and stuff that teaches you about AI:
“Claude 3.7 is More Significant than its Name Implies” [VIDEO] - solid deep dive by AI Explained.
“How I use LLMs” [VIDEO] - a fantastic, info-packed walkthrough by the great Andrej Karpathy.
🤦♂️ AI fail of the week
“Ouch! Burbep is totally getting his tomato pepper ass kicked.”
💰 Sunday Bonus #49: Quickly find relevant Why Try AI guides with my Custom GPT
Next week, my Sunday Bonus + Sunday Showdown archive for paid subscribers will reach 50 entries.
On top of that, I’ve shared dozens of other tips and guides over the years.
Awesome!
But this also means it’s become difficult for readers to find relevant Why Try AI entries that might solve their problem.
So I put together a custom GPT that digs up relevant tips, guides, tools, etc. based on your needs.
Just tell it what you’re after and it’ll give you direct links to relevant entries.